Partial Pose Estimation Of Rigid Object System Using Cad Database

• Main Author: Zulkoffli, Zuliani
• Format: Thesis
• Language: English
• Published: 2017
• Subjects: T Technology; TJ1-1570 Mechanical engineering and machinery
• Online Access: http://eprints.usm.my/46416/1/Partial%20Pose%20Estimation%20Of%20Rigid%20Object%20System%20Using%20Cad%20Database.pdf; http://eprints.usm.my/46416/

Partial pose estimation identification is required for inspection in manufacturing industry. By knowing the partial pose estimation before inspection, overall inspection processing time can be reduced. System development for partial pose estimation identification consists of a few main parts which are image acquisition, pre-processing, processing and camera calibration. Image acquisition is divided into CAD image acquisition and Projection Real Image (PRI) acquisition. Image pre-processing consists of image rescaling, image segmentation and image registration. In image segmentation, high level feature of Outer Box object segmentation method was proposed. In pre-processing development section, the development of CAD model database imitates inspection environment was implemented. The object was represented by the area combined with the edge information of the object. Within this shape representation, partial pose estimation was identified by linking the CAD model database to the inspected object. A few techniques were suggested in pre-processing stage which included 1D Fourier Descriptor, Euclidean Distance, 2D Fourier Descriptor Subspace Matrix and template matching. Partial pose estimation identification using template matching method showed a high performance result. The tested objects were automotive component, bottom automotive component, Arduino board, computer mouse, labelled object and USB connector. Study on template matching was preceded for 360 images CAD model for partial pose estimation identification of ±1 degree accuracy. Ten tests of partial pose estimation were carried out. All the testsshowed the right identification with score 10/10. Average processing time consumption in this ±1 degree accuracy was 1032.6s for automotive component, 997.7s for bottom automotive component, 948.5s for Arduino board, 1198.7s for USB connector and 972.1s for labelled automotive component. For automotive component, partial pose estimation identification linked to random CAD model database gave 974.0s average processing time for ten trials. Then, partial pose estimation identification linked to various CAD models data was studied. Three tests for every object were carried out and gained 3/3 score for automotive component, 3/3 score for bottom automotive component, 3/3 score for Arduino board and 3/3 score for USB connector. After that, study on surface inspection was carried out. Stereo image registration through centre hypotenuse length of outer box method was implemented. The similarity of both CAD stereo image registration and real object stereo image registration resulted in the range of 81.9% to 91.8%.